demowatch | empirical analysis , using the Occupy Movement

demowatch is a Python library. demowatch has no bugs, it has no vulnerabilities and it has low support. However demowatch build file is not available. You can download it from GitHub.

DemoWatch is a way to generate the data and enable the empirical analysis, using the Occupy Movement as a natural experiment and data source. News articles are the proxy by which data can be gleaned from these events, which are by nature difficult to quantify. TagWorks, formerly known as TextThresher, is a platform for easily crowdsourcing the tasks of annotating and parsing text data, developed with DemoWatch and adjacent projects in mind. The backend involves an algorithm of inter-annotator agreement (IAA) to ensure high-quality “consensus” information is generated. Both the TagWorks interface and IAA algorithm were developed by separate Goodly Labs divisions, not DemoWatch researchers. TagWorks works primarily by having annotators answer questions and sometimes highlight text to justify their answers. A specific, contained set of questions and answers is called a schema. A significant amount of sociological theory is fit into a few dozen questions and their respective answers. The questions are presented in a hierarchical fashion, where specific answers to a question may prompt follow-up questions. There is a whole section of the DemoWatch project that focuses on schema development--Nick Adams and Alex Barnard have made enormous contributions here. This involves having a dedicated team of well-trained volunteers user-test the schema, file bug reports, and work on iterating the schema. How can you tell if the event being described in one news article is the same event being described in another news article? Some things may pop out: place, date, number of people, etc. Humans are particularly good at answering this question because our brains can naturally understand the higher-order meanings in the text. For a text parser, the task is much harder. This is especially relevant for us because we’re interested in understanding events, but all we have are news texts. The texts do not have a convenient tag attached to them denoting a unique event ID, which we need to access. We can approximate the unique events through a process called canonicalization. This is the work that Sidney, Jacob, Aaron, Schuyler, Karen, Avik, and Devesh have developed. See the subsection below for more details. Finally, after our canonicalization algorithm has generated a canonical set of events (with associated actions, features, and timesteps), we can perform a quantitative analysis of police-protester interactions. This portion of the project is continually being expanded. Thus far, we have focused primarily on modelling interactions within and among events using regression analysis and dynamic Bayesian networks (probabilistic models).